A hybrid deep learning approach using XceptionNet and vision transformer for accurate chest disease detection from X-ray images

Abstract

Chest radiography is a crucial imaging method in radiology to examine the chest cavity. These images are utilized to identify a range of diseases, such as lung cancer, pneumonia, covid-19, and tuberculosis. Though it is a basic and affordable tool, the accuracy of the interpretation relies heavily on the expertise of the radiologist. In remote and deprived healthcare settings, access to expert medical professionals is a major challenge. Although deep learning-based approaches have been widely exploited in the literature for chest X-ray analysis, they are limited in capturing the full context and long-range dependencies of the entire image. In this innovative research, a new hybrid vision transformer (ViT) architecture known as X-Vision is developed to accurately differentiate between covid-19, normal, and pneumonia from chest X-ray images. The key innovation lies in how features from XceptionNet and ViT are utilized: repeatedly integrated throughout the model’s processing pipeline. By merging the advantages of both the models, this approach had the potential to create more reliable and precise image classification systems. The findings indicate that X-Vision surpasses pre-trained models and attains an average accuracy of 99% for three different classes.